Abstract:

Objective To construct a decision tree model based on MIMIC-Ⅳ database data to evaluate the condition and prognosis of patients with Staphylococcus aureus bloodstream infection（SABI） in intensive care unit（ICU）. Methods The demographic information of 1 030 patients with SABI in ICU and related laboratory indicators were extracted from MIMIC-Ⅳ database. Whether the patient died in hospital was the main outcome. The included patients were classified into training set（773 cases） and validation set（257 cases） in a ratio of 3:1. The random forest model was constructed based on the training set data，and the variables are screened according to the importance score to construct the decision tree model. The model was included in the validation set for validation. Results The importance scores of serum potassium ion（K⁺），serum sodium ion（Na⁺），creatinine（Cr），anion gap（AG），serum calcium ion（Ca²⁺），hematocrit（HCT），hemoglobin（Hb） and international normalized ratio（INR） were high. The decision tree model was constructed based on these 8 indexes. The area under curve（AUC） of receiver operating characteristic（ROC） curve of this model for predicting prognosis of SABI patients （mortality）in ICU was 0.769 4. Conclusions A decision tree model has been constructed to evaluate the condition and prognosis of SABI patients in ICU，which is helpful to rapidly identify the high-risk mortality in this type of patients，which suggests the clinical adjustment of treatment plan in time to improve the prognosis of patients.

Key words: Staphylococcus aureus, Bloodstream infection, Intensive care unit, Prognosis, Decision tree model, MIMIC-Ⅳ database